Optimizing Feature Space of Support Vector Machines Based on Simulated Annealing Algorithm to Classify Polarimetric Images

Considering the development of Remote Sensing sensors, Polarimetric images have been the matter of interest as powerful and efficient tools; since they provide highly informative sources of terrain targets utilizing different polarizations of electromagnetic waves. Classification of a Polarimetric image is an important step to extract such information. Among the several classifiers,Support Vector Machines (SVMs), for their operation based on geometrical characteristics, are particularly more attractive cases in Polarimetric image classification. Although SVMs classifier has superior performance in high dimensional search spaces, redundant or irrelevant features have significantly affect the classification accuracy and the speed of convergence. Therefore, determination of optimum features to achieve an accurate SVMs classifier is an important challenge.Since traditional optimization techniques usually have computational complexities and trap in local optimum in a large search space,Meta-heuristic Algorithms which perform exploration and exploitation to obtain global optimum are applied in this research. Thus,the potentiality of Simulated Annealing Algorithm as a powerful optimization technique in determining the optimal subset of polarimetic features is evaluated in this paper. Also, Genetic Algorithm as a classicmethod is used to be compared with Simulated Annealing Algorithm. The results demonstrate the superior performance of Simulated Annealing Algorithm in terms of classification accuracy and speed of convergence in comparison with Genetic Algorithm.